Defining brain-based OCD patient profiles using task-based fMRI and unsupervised machine learning.

Area of Science:

• Neuroscience
• Psychiatry
• Machine Learning

Background:

• Obsessive compulsive disorder (OCD) research often overlooks neural heterogeneity, relying on group averages that may obscure important patient subgroups.
• Conventional neuroimaging approaches can increase variability and lead to discrepancies in findings due to assumed population homogeneity.

Purpose of the Study:

• To apply unsupervised machine learning to identify distinct neural activity clusters in patients with OCD.
• To investigate heterogeneity in brain network activation, focusing on cognitive control and performance monitoring circuits implicated in OCD.

Main Methods:

• Utilized task-based functional magnetic resonance imaging (fMRI) on 128 adult and adolescent OCD patients and 64 controls.
• Employed unsupervised machine learning to cluster patients based on brain activation patterns in the frontoparietal, cingulo-opercular, and default mode networks.
• Assessed neural activity during cognitive interference and error processing tasks.

Main Results:

• Identified three distinct patient clusters: a 'normative' group (65.9%) similar to controls, an 'interference hyperactivity' group (15.2%), and an 'error hyperactivity' group (18.9%).
• The 'interference hyperactivity' cluster exhibited significantly longer reaction times compared to other patient clusters.
• No other significant demographic or clinical differences were detected between the identified clusters post-hoc.

Conclusions:

• Unsupervised machine learning can reveal distinct neural subgroups within the OCD population.
• These findings challenge the assumption of homogeneity in OCD neuroimaging research and provide a more precise patient characterization.
• This subgroup identification offers a foundation for developing neuroimaging-guided, personalized treatment strategies for OCD.